Skip to main content

Table 5 Overall summary of three clustering techniques

From: A comparison of three clustering methods for finding subgroups in MRI, SMS or clinical data: SPSS TwoStep Cluster analysis, Latent Gold and SNOB

 

TwoStep

Latent Gold

SNOB

Method

Distance-based, agglomerative hierarchical cluster analysis

Finite mixture modeling to probabilistically identify latent classes

Finite mixture modeling to probabilistically identify latent classes

Stopping rule to identify number of subgroups

Automated using either ‘Bayesian information criterion’ or ‘Akaike’s information criterion’

Analyst choice using various criteria, including ‘Bayesian information criterion’, unexplained variance, Chi-square p-value

Automated using ‘Minimum message length’ principle

Suitable data types

Ordinal data require recoding as dichotomous or handled as if interval data

All types

All types

Report classification probability of individuals

No

Yes

Yes

Sensitivity to subgroups

Least

Middle

Most

Reproducibility

Very high

Very high

Very high

Accuracy

Very high

Very high

Very high

Cost

Most expensive

Less expensive

Free

Support

Extensive documentation, fee-based support available

Extensive documentation and some free support available

Some documentation but minimal support available

Interpretability of presentation of results

Results are presented numerically and graphically (charts of certainty of the subgroup structure, bar and pie charts of cluster frequencies, and charts displaying the importance of specific variables to subgroups)

Results are presented numerically and graphically (including a tri-plot displaying the relationships between subgroups)

Results are mostly numeric (although a tree diagram is produced showing the relationship between ‘mother’ and ‘daughter’ subgroups)

Learning curve (subjective judgement)

Easy

Middle

Hard